Abstract
<title>Abstract</title> <p>Background Liver fibrosis is a major contributor to liver-related morbidity and mortality among individuals with previous or current hepatitis B virus (HBV) exposure. Nutritional and inflammatory factors have been implicated in liver fibrosis, however their combined value for fibrosis stage classification remains insufficiently explored. This study aimed to investigate the associations of nutritional and inflammatory indicators with liver fibrosis severity and to develop machine-learning models for fibrosis stage classification. Methods This cross-sectional study analyzed data from the National Health and Nutrition Examination Survey (NHANES) 2017–2018 and 2019–2020 cycles. Individuals with evidence of previous or current HBV exposure (anti-HBc positive) were included. Liver fibrosis severity was assessed using transient elastography-derived liver stiffness measurements. Associations between clinical variables and fibrosis severity were evaluated using correlation analysis, univariate regression, and restricted cubic spline models. Random forest-based feature selection was performed, and six machine-learning algorithms, including Random Forest Support Vector Machine, Extreme Gradient Boosting (XGBoost), K-Nearest Neighbor, Decision Tree, and Neural Network, were developed and evaluated using repeated 10-fold cross-validation. Model performance was assessed using accuracy, multi area of under curve, F1-score, calibration analyses and . Results A total of 911 participants were included, comprising 769 individuals without fibrosis (84.41%), 39 with F1 fibrosis (4.28%), 52 with F2 fibrosis (5.71%), and 51 with F3 fibrosis (6.00%). Correlation and regression analyses demonstrated significant associations between liver fibrosis severity and multiple nutritional and inflammatory indicators. Restricted cubic spline analyses further revealed nonlinear relationships between fibrosis severity and age, BMI, GNRI, NPAR, and SIRI. Feature selection identified 10 key variables, including GNRI, BMI, albumin, total cholesterol, CRP, NPAR, creatinine, HDL cholesterol, waist-to-height ratio, and age. Among the six machine-learning algorithms evaluated, XGBoost achieved the best overall performance, with an accuracy of 0.940, a MAUC of 0.940, and a macro-F1 score of 0.804. Conclusions Nutritional and inflammatory indicators were significantly associated with liver fibrosis severity among individuals with previous or current HBV exposure. An XGBoost model constructed from routinely available clinical variables showed good performance in fibrosis stage classification and warrants further validation in independent cohorts.</p>